Towards an Auto-nowcasting System for Landslide Hazards

Cheng-Chien LIU, Hsiao-Yuan YIN, Hsiao-Wei CHUNG, Wei LUO and Ke-Wei YAN

This paper reports the efforts we made in the past four years to develop an auto-nowcasting system of landslide hazards, as well as the accuracy assessment of the I-Lan case in 2012 and 2013. This system uses a nondeterministic geometric mean model of landslide susceptibility index (LSI) to integrate three grid-based preparatory factors: slope, aspect, lithology, as well as one region-based factor: total flux. The weight of each preparatory factor is calculated by excluding the union of all shaded areas in the time series of satellite observations. The coefficients of the nondeterministic geometric mean model are determined by maximizing the positive right and negative right regions, and minimizing the missing and false alarm regions. By assuming that landslide hazard index (LHI) is related to LSI and accumulated precipitation (AP), the event-based inventory and AP for Typhoon Saola (11 July 2012) and Typhoon Soulik (12 July 2013) are used to develop and validate the LHI of I-Lan. Results show that the overall accuracy is as high as 85%. This LHI model has been implemented through SWCB Sediment-related Disaster Geospatial Information System (SDGIS) to provide an auto-nowcasting service with a map of five-level warning updated every hour.

2018/1 336-340